mpfaffenberger/code_puppy

What it solves

Code Puppy is an AI-powered code generation agent designed to replace expensive IDEs by providing a high-quality, open-source alternative for programming tasks. It allows developers to use a wide variety of AI models to generate code, manage files, and execute shell commands directly from a terminal-based interface.

How it works

The tool operates as a CLI agent that integrates with numerous LLM providers (via models.dev, OpenAI, Anthropic, Google, etc.) and supports the Model Context Protocol (MCP) for external tools. It features a "Speculative Tool Execution" mode that runs tool calls while the model is still typing them to reduce latency. It also supports durable execution via DBOS, allowing agent runs to survive crashes and be resumed.

Who it’s for

Developers who want a powerful, model-agnostic AI coding assistant that is more flexible and affordable than proprietary AI IDEs, and those who prefer working in a terminal environment.

Highlights

• Speculative Tool Execution: Reduces latency by executing tool calls (like reading files or grepping) as they are streamed from the LLM.
• Extensive Model Support: Integration with models.dev provides access to 65+ providers and over 1,000 models.
• Durable Execution: Optional DBOS integration for checkpointing interactions and recovering from crashes.
• Customizable Agents: Ability to create specialized AI assistants via Python or JSON configurations.
• Round Robin Distribution: Distributes requests across multiple API keys to bypass rate limits.
• Agent Rules: Supports AGENTS.md files for project-specific coding standards and behavioral guidelines.